The Stable Gallup Ballot

The Stable Gallup Ballot

By Sean Trende - September 21, 2010

I've done the usual writeup of the "latest polls," but we'll run it this afternoon. I thought a short piece on the Gallup ballot was in order for the morning. Conservatives and liberals alike have been complaining about how "bouncy" Gallup is -- one week it has the Republicans up by an unprecedented 10 points, the next it shows a tie. 

And so, we can expect some buzz now that it is showing Republicans down a point, 46 percent to 45 percent.

First, some perspective. It is highly unusual for Republicans even to be tied or down one in the generic ballot, especially in the registered voter model. This result is probably pointing to R+4 to R+6 in the general electorate, which would be a tsunami on par with 1994.

But more importantly, since February of this year, Gallup has shown Democrats between 43 percent and 47 percent with only two exceptions. Since February of this year, Gallup has shown Republicans between 44 percent and 50 percent, again with two exceptions.

In other words, this is what we'd expect to see if Democrats were *actually* pulling in around 45 percent of registered voters, and the GOP was at 48 or so. Occasionally we'd see a big Republican lead, occasionally we'd see a tie, or even a small Democratic lead -- we might even see a one-week swing from a seven-point GOP lead to a 3-point Democratic lead! 

That's what that error margin that pundits keep talking about is all about -- you don't hit the actual numbers all that often, but you get close to them.  Even looking at Gallup's gigantic samples for September and August of some 7,000 registered voters, the movement of Republicans from 49 points to 46 points and of Democrats from 43 points to 45 points is pretty close to the error margin we'd expect to see in with 7,000 voters (+/- 1.2 percent or so).

Why don't we see this with other pollsters? I suspect that they weight more heavily than Gallup does. This is especially true of Rasmussen Reports, which uses what I consider to be an ingenious method of weighting by partisan ID by using the identification from other polls to set the registration numbers. This has the effect of tamping down on random variance, by reducing the number of outliers that sampling will produce.

So I think what is going on is that we're so used to getting results from pollsters who weight heavily that we forget what things would look like from a pollster that isn't employing a bunch of screens. Again, I'm not saying one method is better than the other. I'm just saying that when you see something "odd" in a poll, it typically just means that the pollster is employing a methodology that is different from what others are using.

Sean Trende is senior elections analyst for RealClearPolitics. He is a co-author of the 2014 Almanac of American Politics and author of The Lost Majority. He can be reached at Follow him on Twitter @SeanTrende.

Latest On Twitter